Gemini for healthcare is Google Cloud's agent-based industry offering for health systems and health plans, announced on November 10, 2026 and available in preview in the US. It gives staff skills for tasks such as medical records synthesis and coverage review, connects read-only to systems such as Epic and Oracle Health, and lets an organization choose among models from Google, Anthropic and OpenAI. Before anyone turns it on, or turns on ChatGPT or Claude for Healthcare, the practical work is the same: signed agreement, mapped data flows, access controls, audit logs and human review.
Who this is for and what you will have at the end
This guide is for IT leads, security staff and practice or operations managers in US healthcare organizations that are being asked, "Can we try Gemini for healthcare?" It is not medical advice, and it makes no claim about clinical accuracy. Nothing here tells you which tool is best.
By the end you will know what Google announced and what it left unstated, who can get access, what the three vendors say about HIPAA, data use and connectors, and a checklist you can run before any pilot. All vendor statements below are attributed to the vendor and were checked on 2026-11-20.
What Google announced
Google's blog post, written by Thomas Kurian and dated November 10, 2026, says Gemini for healthcare "extends the universal Gemini agent" introduced the month before at its Gemini at Work event. It names providers (health systems, medical groups) and payers (health plans) as the audience. Example work includes chart preparation, authorization packets, payer policy checks and member benefit explanations.
The post describes four parts:
- Skills. Reusable instructions and context written by domain experts that teach Gemini a specific task. Named skills include medical records synthesis, pre-appointment preparation, coverage and policy review, and payer benefit determination. Teams can also build custom skills around their own procedures.
- Connectors. A library that links Gemini to EHR, claims, care management and administrative systems.
- Partner ecosystem. Systems integrators and digital health partners.
- A governed platform with open model choice. Google says it is "built to support HIPAA compliance, with enterprise-grade access controls, end-to-end audit logging, and verifiable citations."
Availability and pricing
Google's blog says it is available in preview in the US. The press release adds that it launched "initially available in preview" to Gemini Enterprise Standard and Gemini Enterprise Plus customers, with per-user seat subscriptions or pay-as-you-go. Neither page we opened lists prices. If you do not already have a Gemini Enterprise edition, the practical first step is a conversation with your Google Cloud account team, not a sign-up form. For the broader Gemini plan lineup see Gemini in Google Workspace plans.
What it connects to
The two connectors that matter most to providers are the EHRs. The blog says the Epic connector gives authorized users read-only access to patient conditions, medications, labs, clinical notes and specialist documentation, "with no write-back to the underlying chart." The Oracle Health connector is also read-only, with "zero write-back." The press release says Epic access is via open.epic APIs and Oracle Health access is via Ignite APIs for FHIR, and that a NextGen Healthcare connector is planned "in the coming months."
Other connectors named by Google include InterSystems FHIR Server, Verato, Keragon, Waystar, GoodRx, Infinitus Relay, IMO Health, ICD-10 codes, RxNorm, CMS Coverage Data, the NPI Registry, Optum EvidenceVue, BMJ Digital Health, Salesforce Health Cloud, ServiceNow, Microsoft 365 and Google Workspace. Optum Network Eligibility and Coordination of Benefits is described as in "limited release and testing", so do not assume every listed item is generally usable.
Model choice
The press release says the platform supports "open model choice across Google, Anthropic, and OpenAI," routing high-volume administrative work to fast, low-cost models and complex analysis to high-reasoning models. For governance, this means one tool may send a prompt to a model from a different vendor than the one you contracted with. Ask Google which models are enabled by default in your tenant and how the agreement treats each.
Data and access statements
Google states that clinical data, prompts and generated outputs "remain within your organization's governed environment and are never used to train or fine-tune Google's foundation models." The press release says each agent has a distinct identity and "inherits each user's existing role-based permissions," and that connectors preserve existing role-based access controls. These are vendor claims; verify them in a pilot rather than assuming them.
Named launch customers
Google names HCA Healthcare, Highmark Health, Northwell Health, Optum, Blue Cross Blue Shield of Massachusetts, Hartford HealthCare and Stanford School of Medicine. It lists the American Society of Clinical Oncology as a contributor. Being named does not mean those organizations use it for all staff or in production, and Google's pages do not say that.
How it compares to ChatGPT and Claude for Healthcare
Both competing products launched in January 2026. The table includes only what each vendor's pages said when we checked; a blank means we did not find it stated, not that it does not exist.
| Question | Gemini for healthcare (Google) | ChatGPT for Healthcare (OpenAI) | Claude for Healthcare (Anthropic) |
|---|---|---|---|
| Announced | Nov 10, 2026 | Jan 8, 2026 | Jan 11, 2026 |
| Status | Preview, US | Per OpenAI help pages, listed as HIPAA eligible with a BAA | New connectors and Agent Skills described as generally available to Claude subscribers |
| HIPAA wording | "Built to support HIPAA compliance" | BAA required to use with PHI; BAA covers only listed products and features | "HIPAA-ready products" for providers and payers; BAA not mentioned on the pages we opened |
| Data used for training | Google says never for foundation models | OpenAI says content shared is not used to train models | Anthropic says "We do not use users' health data to train models" |
| Healthcare connectors | Epic, Oracle Health (read-only), InterSystems FHIR, many more | Not verified on OpenAI's own page this run | CMS Coverage Database, ICD-10, NPI Registry, PubMed; a FHIR skill |
| Model choice | Google, Anthropic, OpenAI | Not verified | Claude models |
| Who gets it | Gemini Enterprise Standard and Plus | Eligible customers can apply for a BAA | Claude for Enterprise and the Claude Platform for healthcare customers |
Read the table as a prompt for questions, not a ranking. The three vendors describe different things: Google sells an agent platform with deep EHR connectors, Anthropic emphasizes connectors to reference data and a FHIR skill, and OpenAI's help pages emphasize the scope of the BAA. The OpenAI details came from OpenAI's own help-center and product pages as surfaced in search results; OpenAI's servers blocked our page fetches, so confirm them on OpenAI's site. For a pricing view of the general business tiers, see ChatGPT Team vs Claude Team vs Gemini Business.
Security and compliance checklist before any pilot
Use this list for any of the three. It does not replace advice from your compliance officer or counsel.
1. Agreement first
- Confirm in writing that a business associate agreement (BAA) is in place with the right legal entity before any protected health information (PHI) enters a prompt, file or connector.
- Read which products and features the agreement covers. OpenAI's pages say the BAA covers only listed functionality, and some features sit outside it and are disabled by default. Ask each vendor for the equivalent list.
- Ask whether preview status changes terms. Google describes Gemini for healthcare as a preview, and preview services often have different support and service terms. Ask for the answer in writing.
2. Map the data flow
Draw what leaves each system: which EHR fields a connector reads, where prompts are processed, where outputs and logs are stored, and whether another model provider sees the prompt. Read-only EHR access still exposes PHI to the agent, so "no write-back" reduces one risk, not all of them.
3. Access control
- Test that the agent only returns what the signed-in user could already see. Google says agents inherit each user's role-based permissions; check with a low-privilege test account.
- Decide who can create custom skills, because a skill is effectively a reusable instruction set that touches records.
- Turn off or restrict features you have not reviewed, such as memory or scheduled tasks, until reviewed.
4. Audit logs and retention
Confirm you can export logs of who asked what and which records were retrieved, and for how long. Google lists "end-to-end audit logging"; confirm the log format, retention and export path with your account team.
5. Human review
Define who reviews outputs before they affect a patient, a claim or a coverage decision. Require citations to the source record or policy, and make it clear in the workflow that a person owns the final decision.
6. Start narrow
Pilot one low-risk administrative workflow, such as policy lookup that needs no PHI, with a named owner and a stop condition before moving to chart-related work.
Troubleshooting common stumbling blocks
| Problem | Likely cause | What to do |
|---|---|---|
| Option not visible in your Gemini admin console | Preview limited to US and to Enterprise Standard or Plus customers | Check your edition and region, then ask your Google Cloud team about preview enrollment |
| EHR connector returns nothing | EHR-side approval or API access for the app has not been set up | Ask your EHR administrator what API access Epic or Oracle Health requires for your site; the vendors' pages do not describe this setup |
| Vendor says "HIPAA-ready" but gives no BAA | Marketing wording is not an agreement | Stop and request the BAA and a scope list before any PHI use |
| A listed connector is unavailable | Some items are in limited release or planned | Treat Google's list as a roadmap plus availability, and ask for the current status per connector |
What we checked
- Opened Google Cloud's "Introducing Gemini for healthcare" blog post on 2026-11-20; it is dated November 10, 2026 and describes a US preview, read-only Epic and Oracle Health connectors and the "never used to train or fine-tune" statement.
- Opened Google Cloud's press release on 2026-11-20 and cross-checked it against the blog: the press release adds Gemini Enterprise Standard and Plus eligibility, per-user or pay-as-you-go options, NextGen "in the coming months," and the open.epic and Ignite FHIR API wording. The two pages list slightly different launch customers (Hartford HealthCare and Stanford School of Medicine appear only in the press release), so we used the press release list and named the source.
- Opened Anthropic's healthcare announcement (dated January 11, 2026) and the Claude healthcare page on 2026-11-20. Neither mentioned a BAA, so the table says so.
- OpenAI's pages returned HTTP 403 to our fetch tool. The OpenAI column relies on OpenAI-domain pages as summarized in search results on 2026-11-20 (January 8, 2026 launch, BAA required, scope limits). We did not open the pages themselves.
- Google's product page (cloud.google.com/gemini/healthcare) returned no body text, so pricing and terms there were not checked.
- Not checked: we did not use any of the three products, and we did not test connectors, logs or access controls on a live tenant. Everything above follows vendor documentation and announcements as of 2026-11-20.
Last tested: 2026-11-20
References
- Google Cloud blog, Introducing Gemini for healthcare: https://cloud.google.com/blog/products/ai-machine-learning/gemini-for-healthcare
- Google Cloud press release, November 10, 2026: https://www.googlecloudpresscorner.com/2026-11-10-Google-Cloud-Launches-Gemini-for-Healthcare
- Google Cloud, AI for healthcare: https://cloud.google.com/gemini/healthcare
- Anthropic, Advancing Claude in healthcare and the life sciences: https://www.anthropic.com/news/healthcare-life-sciences
- Anthropic, Claude for healthcare: https://claude.com/solutions/healthcare
- OpenAI, OpenAI for Healthcare: https://openai.com/index/openai-for-healthcare/
- OpenAI Help Center, ChatGPT for Healthcare: https://help.openai.com/en/articles/20001046-chatgpt-for-healthcare
- OpenAI Help Center, HIPAA eligible products and functionality: https://help.openai.com/articles/20001069
Comments
Questions and tips are welcome. Comments are checked by the EpiFive team before they appear; please do not share passwords, phone numbers or other private details.